Messages in this thread | | | From | (H. Peter Anvin) | Subject | Re: CONFIG_RANDOM option for 1.99.2 | Date | 31 May 1996 01:47:25 GMT |
| |
In article <Pine.OSF.3.91.960517093442.5381G-100000@namu23.gwdg.de>, Martin.Dalecki <dalecki@namu23.Num.Math.Uni-Goettingen.de> wrote: > > No definitly false. Monte Carlo methods are based on *equally distributed* > sequences. They are not based on *random* sequences. That's a subtile > difference! >
It is also a completely bogus assertion. The sequence 1, 2, 3, 4... , 100, 1, 2... is equally distributed on the integers [1,100] but is useless for MC. MC requires *uncorrelated* numbers. That is *not* a subtle difference!
Equal distribution is not a necessity, however. It can easily be shown that as long as the uncorrelated criterion is satisfied, a biased random number source can be turned into an unbiased source with lower production.
> And finally random.c is not as random as You may beleve. Start it on an > otherwise not busy machine to see why! Yust do cat /dev/random and see > the random numbers coming precisely after any keybord hit. This isn't > acceptable for any number cruching, since those are mostly programms > which are supposed to run in core and on otherwise idle machines.
random.c is just fine and random, but it doesn't have unlimited output without special hardware. That is why there is /dev/urandom.
-hpa
| |